Tesla beefs up semi-autonomous suite with Nvidia’s parallel computing platform

Bruce Brown
Tesla beefs up semi-autonomous suite with Nvidia’s parallel computing platform
Tesla's 'Vision' program will use an Nvidia GPU with the CUDA parallel computing platform. The end result will be an end-to-end system that can take data from Tesla's radar and sensor suite and process the incoming image data to control vehicles.

It appears that Tesla and Nvidia are taking huge steps forward together in building the next version of Tesla’s semi-autonomous driving system. The new Tesla Vision incorporates Nvidia’s CUDA parallel computing platform, according to Electrek.

Tesla previously used a vision system from Mobileye for its Autopilot semi-autonomous platform. That relationship ended in terms of ongoing development during the summer, following a widely publicized fatality in a Tesla S when neither the Autopilot system nor the human driver noticed a truck crossing the road from the left.

Related: Nvidia offers a peek under the hood of its Drive PX 2 platform for self-driving cars

At the time the split with Mobileye was announced, many assumed Tesla would develop its own vision system. CEO Elon Musk believes in bringing component development, manufacturing, and control in-house whenever practical and profitable, so creating its own Vision component made sense.

Now the story is, according to Electrek sources close to the Tesla Vision program, that rather than developing a vision module system on a chip (SoC), vehicle autonomous features will be supported by a full parallel computing platform based on Nvidia’s CUDA platform. Tesla has not confirmed or announced details about the system, but it sounds like Nvidia graphics processing units (GPUs) will be used to process the many images captured by the Tesla’s growing suite of sensors, cameras, and radar.

Incorporating more Nvidia components in Tesla’s Autopilot hardware makes sense for several reasons. Tesla already uses Nvidia processors in media control units and instrument clusters for the Model S and X.  Even more strategic are Nvidia’s several autonomous and semi-autonomous driving platforms. The latest, the Nvidia Drive PXd 2 is a liquid-cooled, 12-core platform with a Pascal GPU. Referred to as the “World’s First AI Supercomputer for Self-Driving Cars,” the PX 2 is sufficient to fully support a Level 4 self-driving car. Level 4 refers to a vehicle that is capable of driving by itself with no human intervention under some conditions.

Another factor in the Nvidia Tesla alliance may be the fact that Nvidia CEO Jen-Hsun Huang is a huge Tesla fan, owning several Tesla models himself.